Quantitative comparison of spatial fields for hydrological model assessment - some promising approaches

The current practice for assessing spatial predictions from distributed hydrological models is simplistic, with visual inspection and occasional point observations generally used for model assessment. With the increasing availability of spatial observations from remote sensing and intensive field studies, the current methods for assessing the spatial component of model predictions need to advance. This paper emphasises the role that spatial field comparisons can play in model assessment. A review of the current methods used in hydrology, and other disciplines where spatial field comparisons are widely used, reveals some promising methods for quantitatively comparing spatial fields. These promising approaches-segmentation, importance maps, fuzzy comparison and multiscale comparison-are for local comparison of spatial fields. They address some of the weaknesses with the current approaches to spatial field comparison used in hydrological modelling and, in doing so, emulate some aspects of human visual comparison. The potential of these approaches for assessing spatial predictions and understanding model performance is illustrated with a simple example. (C) 2004 Elsevier Ltd. All rights reserved.